Traditional Software vs AI Agents
Same input. Different output. Every time.
Traditional Software
Deterministic
Input → Output
Same input always produces the same output. Predictable by design.
Testing
Unit tests verify boolean conditions. Pass = correct.
Failure mode
Code either works or it crashes. The boundary is clear.
vs
AI Agents
Probabilistic
Input → Output
Same input can produce different outputs. Distribution of answers, not a single value.
Testing
Unit tests can pass while the agent answers incorrectly in production.
Failure mode
The agent runs, responds confidently, and may still be wrong. Silent failure.